Re: [PATCH 07/11] xfs: superblock scrub should use uncached buffers

[Dave Chinner <david@fromorbit.com>]

On Thu, Apr 19, 2018 at 10:25:48AM -0700, Darrick J. Wong wrote:
> On Thu, Apr 19, 2018 at 08:55:47AM -0400, Brian Foster wrote:
> > On Tue, Apr 17, 2018 at 07:40:20PM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > 
> > > We've never cached buffers when reading or writing superblocks, so we
> > > need to change scrub to do likewise or risk screwing up the uncached sb
> > > buffer usage everywhere else.
> > > 
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > 
> > Hmm, so this goes back to this[1] thread, right?
> 
> Yep.
> 
> > IIRC, we wanted to end up with an lru-bypassing uncached buffer lookup
> > mechanism to provide uncached behavior for resource-saving purposes
> > but without introducing serialization issues between multiple users of
> > uncached buffers.
> 
> Right.
> 
> > On a quick look back, growfs currently uses cached buffers for secondary
> > superblocks and the associated patch was looking to change that to
> > something like the above. Don't we have the same requirement here (since
> > growfs currently still uses cached buffers)?
> 
> Correct, this patch is contingent on landing Dave's "tableise growfs"
> series that converts growfs to use uncached buffers for writing out the
> new secondary supers.  If I manage to land Dave's growfs thing before
> repair then I'll keep this; if repair goes first I'll defer it to that
> series.

I plan on changing that patch set to use single-use cached buffers
rather than uncached buffers. So I'd just drop this patch, or
replace it with a patch that adds this:

#define XFS_SINGLE_USE_REF	0

and this after the xfs_trans_read_buf() call:

	xfs_buf_set_ref(bp, XFS_SINGLE_USE_REF);

Which will result in it being reclaimed when it is released rather
than being put on the LRU.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com